A stereo vision tracking algorithm based on 3-dimensional delaunay triangulation
Liu Guo-dong, Sha Liu, Jianxun Li · 2016
Single-camera object detection makes three-dimensional information into two-dimensional images, which misses the three-dimensional structure information of object. In addition, due to the changes in perspective, scale and attitude, it is difficult to track object robustly. In this paper, we put forward a stereo vison tracking algorithm based on 3-dimensional (3D) Delaunay triangulation. Firstly, 3D key points with Scale Invariant Feature Transform (SIFT) descriptors are extracted in 3D point clouds. Then, 3D Delaunay triangulation is built to represent the spatial configuration information for the feature points. Based on the unique topological structure of 3D Delaunay triangulation of a set of key points, the object representation is obtained. Finally in particle filter framework, 3D target tracking is in process. Here, we fully exploit spatial geometry to restrict the feature points, which build a discriminative model for object tracking. Experiments show that the method has good tolerance for changes in illumination, scale and attitude.